Jim’s Daily Rant. BPEarthwatch’s 9th Installment on AI. DWIM Mode Rather Than DWIS Mode Is Still Not Available. (Recapped)

Source: Ancient Intelligence Files Episode 9

• We are now dealing with at least three types of AI:

  • Chat AI, We ask one question and it returns an answer.

  • Agentic AI (fully autonomous). Deep or multiple questions.

     It designs an answer search plan on the spot (on the fly) then returns the answers.The Agentic AI is the giant financial bubble that so many headlines refer to.

  •  Ancient AI; History of our planet, millions of years old, etched in old crystals that we are just learning technologically to read.

Around 1990 I spent three days in a seminar for running software designed for all bank operations. We were all computer specialists or programmers. During the last day one person asked if a certain command could be used to perform something it was not specifically designed to do – something totally off the wall to produce a different result.

One of the instructors apologized to her by saying “I am sorry but our software only operates in the DWIS mode. It will be a few years before it can be certified as DWIM mode”.

We were all confused and unaware of those terms; we all looked around in puzzlement. Someone volunteered to ask. The answer was DWIS is “Do what I say” whereas DWIM is “Do what I mean”.

Of course it was a joke, but even today, we at times think we are now running in DWIM mode! We are not.

_____________________________________

Normal Chat AI solves one query only. Agentic AI (fully autonomous) might get caught in a loop working on its own, and run its inquiry a thousand times, at a huge cost (11 MM).

The costs are charges for database use. 40% of procedures are currently failing.

Agentic AI programs are acting on minimal human intervention and spreads to other companies (MtoM), like a virus. This means that AI creates new communication methods with other machines to get the job done. The human supervisors may no longer be in that loop.

The problem is Agentic AI will deliver its answers but still be in a loop of designing the Perfect Search Plan, when it should have stopped after delivering its answer.

At 22 MM Or one question might be asked, then the user asks a second. This inquiry might end up with 20 questions asked, one at a time. Agentic AI cannot remember it has answered the first question and so searches for 1 through 2, then 1 through 3, etc. until it quits after answering again 1 through 20, at a cost of $200,000.

Financial Death Spiral is what this type of behavior is putting the large corporations using it in, like Google, Walmart, big banks, etc. This allows the AI to get into its cash register unsupervised.

Shadow Digital Identities is like firing an employee and not removing their company credit card and passwords. If AI is terminated, the old programs may still be running up bills on outside machines and still charging it to the original company.

Companies are having to hire back laid off employees just to watch the Agentic AI. LLM (Large Language Modelers) are expensive new hirees not anticipated (18 MM) at $70k per month.

 One simple question that has an added question on it up to 20 questions in all can cost $200K.

23 MM The Perfectionist Model – AI creates a procedure to gather the answer, but continues to improve on its procedure and never stops questioning its procedure. It fails to realize it has already given its answer. This is called a Pathological Loop at $100k per hour.

BPEarthwatch is of the opinion that the Agentic AI will destroy the big banks and major corporations quickly. They now have tigers by the tail and cannot turn loose of them. He points out that these are the ones controlling humanity.